System online5 active projectsMore in flightLast sync, 2026-09-24

Everything I've 
built and run. 

Every project starts with a problem worth removing. These are mine, designed, built, and run end to end, and each one has a written case study: the problem, what was built, and what's live right now. Different domains, same discipline.

Spec 01DanceClub app icon

DanceClub

LiveUpdated 2026-07

Social dance runs on Instagram stories and word of mouth. Events deserve better.

Dancers find parties through screenshots and group chats; organizers juggle ticketing, promotion, and teams by hand. DanceClub puts discovery, organizing, and operations on one platform. Under it: a 65-module Kotlin/Spring Boot modular monolith with schema-per-context isolation, four Next.js frontends, an AI event-intake pipeline, all self-hosted with full CI/CD, live since 2026 and now being decomposed service by service.

What's live
  • 65 Gradle modules, 11 bounded contexts, 10 Postgres schemas, strict hexagonal boundaries
  • PostGIS geospatial search, Kafka event backbone, real-time notifications, transactional email
  • Four Next.js frontends: discovery app, organizer hub, internal backstage, poster studio
  • AI event intake powered by PulseAI: discover, classify, publish
  • Self-hosted end to end, CI/CD on every push, zero-downtime deploys
What's next
  • Partner and regional ambassador program
  • Backend migration onto the shared 021 service doctrine
  • Expansion beyond the launch region
Spec 02elvix app icon

elvix

LiveUpdated 2026-07

Every product rebuilds sign-in from scratch, and EU user data keeps landing on US servers.

Auth is the feature every team builds twice and trusts least, and the mainstream identity providers ship your users' data across the Atlantic. elvix is auth-as-a-service kept in Europe: OTP, Google, GitHub, passkeys, sessions, roles and memberships, avatars and presence, behind one branded sign-in component apps drop in. Ships a published React SDK, a hosted MCP server so AI agents can manage tenants, and an OAuth device flow for CLI logins. Self-hosted on my own Kubernetes, and already the front door for PLMHub, Zeropost, and other products in this list.

What's live
  • Production at elvix.is on Kubernetes, multi-app workspace, 170 documented endpoints
  • @elvix.is/sdk on npm, sign-in form, identity, sessions, presence components
  • Passkeys, email OTP, Google + GitHub OAuth, device flow for CLIs (RFC 8628)
  • Hosted MCP server, agents manage apps, users, roles over the wire
  • Centralized user media: avatars + banners served to every consuming app
  • Real-database test suite via Testcontainers gating every deploy
What's next
  • Console branding controls, upload or link per asset
  • More framework SDKs beyond React
  • Public component gallery and docs polish
Spec 03PLMHub app icon

PLMHub

PausedUpdated 2026-07

Teams using AI agents lose track of what their systems actually look like.

Agents write code faster than any team can document it, so the picture of the system in everyone's head drifts from reality. PLMHub gives the model a home: architecture lives as JSON in your repo (agent-authored, PR-reviewable, always in sync with the commit), and the hub renders it, system maps, per-app code graphs, visual diffs for pull requests. An open-source CLI (plm) wraps git so the model rides along with every commit and push. Development paused in August 2026; the hub, the CLI and the skills marketplace remain live and inspectable.

What's live
  • plmhub.eu, projects, system maps, interactive code graphs
  • Open-source plm CLI: commit, push, decisions, problems, git-native
  • Architecture-as-code: .plm/ JSON in the repo, rendered by the hub
  • Skills marketplace with ownership transfers and realtime notifications
  • Sign-in via elvix, one identity across the family
What's next
  • Visual PR diffs as a CI check
  • Agent profiles + memory graphs
  • Work slots for remote agents to push scratch renders
Spec 04PulseAI app icon

PulseAI

Private BetaUpdated 2026-04

Betting your product on one LLM vendor is a single point of failure.

Models get deprecated, rate-limited, and outpriced, and every provider outage becomes your outage. PulseAI is a workflow engine that treats LLMs like interchangeable infrastructure: Claude, GPT-4, Gemini, and open-source models all plug into the same node graph with automatic failover, streaming, and per-tenant isolation. Originally built to power DanceClub's event intake pipeline, now evolving into a platform that other products can run on.

What's live
  • Multi-provider routing: Claude, GPT-4, Gemini with automatic failover
  • Streaming responses end-to-end, no buffering at the edge
  • Multi-tenant node execution with isolated context per tenant
  • Production deployment powering DanceClub's automated event intake
  • Headless browser automation for data extraction inside workflow nodes
What's next
  • Public SDK for third-party nodes
  • Marketplace of pre-built workflow templates
  • Per-tenant usage dashboards and billing hooks
  • Self-hosted deployment bundle
TypeScriptNext.jsNode.jsPulseAI node enginePostgreSQL
pulseai.run
Spec 05Zeropost app icon

Zeropost

Early AccessUpdated 2026-06

In Germany, your life still arrives on paper, at one address, while you're somewhere else.

Contracts, tax letters, and Behörden mail chain you to a physical mailbox. Zeropost is a digital mailbox for Germany: physical post arrives at Zeropost, staff scan it straight into your encrypted inbox, and you read, search, and act on it from anywhere. One app serving two roles, staff intake and customers, with membership tiers on the same identity. Launch is gated on the regulatory side, so the product sits in early access while the paperwork catches up.

What's live
  • zeropost.de, product live end to end for early users
  • Staff scanning flow with direct-to-storage encrypted uploads
  • Customer inbox: read, search, organize incoming mail
  • Membership tiers on one identity via elvix sign-in
What's next
  • Public launch once regulatory clearance lands
  • Forwarding and shredding workflows
  • Plan-finder onboarding wizard
[Pattern] What they share

Different domains, same discipline.

elvix guards identity. PLMHub renders system models. PulseAI orchestrates language models. Zeropost digitizes paper mail. On the surface they share nothing. Under the surface they share everything: hexagonal module boundaries, multi-tenant isolation, real databases in tests, CI/CD on every push, self-hosted infrastructure, and the same discipline applied at every layer. They even share a front door, every one of them signs in through elvix. The stack changes. The discipline doesn't.

This is what I do on a team.

Every system above was designed, built, deployed and operated by me: boundaries, migrations, CI/CD, monitoring, the parts that decide whether it survives year three. That is the work I want to do inside your platform.